In this study, we investigated how learners allocate processing resources when repeating Japanese sentences containing nonsense words by using an articulatory suppression task. Working memory (WM) capacity was manipulated as an individual factor. Nonsense words were placed at the beginning (beginning condition) or in the middle of sentences (middle condition). The results suggested that (a) regardless of WM capacity, participants performed semantic processing while listening to the auditory presentation. (b) During the pause between the end of the auditory presentation and the start of oral reproduction, participants reconstructed sentences using the linguistic knowledge learned in the beginning condition, while they reconstructed sentences using the phonological information of the original sentences in the middle condition. (c) The difference in WM capacity affects the phonological retention of nonsense words. Participants with larger capacity could accurately reproduce nonsense words, while participants with smaller capacity could not. These results indicate that when sentences containing nonsense words are repeated, participants with larger capacity can perform both semantic processing and phonological retention in a well-balanced manner, while participants with smaller capacity can only achieve either semantic processing or phonological retention.